JP6556660B2 - Characteristic understanding device, method, and program - Google Patents

Description

  The present invention relates to a characteristic understanding device, method, and program.

  There is a need to understand the characteristics of roads in order to realize various services such as route search. For example, by understanding the characteristics of which roads are lively and bright and preferentially searching for those roads, route search that meets the needs of users who want to go through busy roads or bright roads is realized. it can.

  There are several existing efforts to understand the diverse characteristics that are geographically distributed in the real world.

  One approach is to manually add information on the characteristics of each road (Non-Patent Document 1). These approaches involve manually evaluating whether or not features specified in advance using crowdsourcing or the like apply to each road, and assigning the result to each road on the road network for use in route search or the like. The road network here is a data format in which a road network is expressed in a graph format with an intersection as a node and a road as an edge, and is generally used for route search and the like.

  Another approach is to estimate the characteristics of each road from statistical information (Non-patent Document 2). It is an effort to estimate the safety level of each road and the amount of green gas emissions based on the open data published by the government and the movement trajectory information such as cars.

  Another approach is to try to understand local information using information on the Web such as social media (Non-patent Document 3). These efforts are intended to understand how the characteristics are distributed in POI (Point of Interest) units, circles, and cell units, based on textual data associated with latitude and longitude. . The POI here is a spot where a person visits with interest such as a store or a sightseeing spot.

Daniele Quercia, Rossano Schifanella and Luca Maria Aiello “The shortest path to happiness: recommending beautiful, quiet, and happy routes in the city”, in Proc. Of Hypertext, 2014. B. Yang et al, “Using incomplete information for complete weight annotation of road networks”, in IEEE Transactions on Knowledge and Data Engineering, 2011. Zhijun Yin et al, “Geographical topic discovery and comparison”, in Proc. Of WWW, 2011.

  However, the conventional techniques related to the above non-patent documents have the following problems.

  The first problem is that the method of manually assigning information and extracting specified information is expensive and cannot be applied to a wide area or acquisition of various information.

  The second problem is that the type of information that can be collected is limited by the approach of estimating characteristics from statistical information, and characteristics relating to human subjectivity cannot be estimated.

  The third problem is that there are efforts to understand characteristics in POI units and area units in efforts to understand local information using information on the Web such as social media, and there are also efforts to assign the results to roads. However, they do not consider the relationship between roads and POIs or areas.

  As a technique for understanding the characteristics of a road by using characteristic understanding in units of POIs, a technique of giving POI information to neighboring roads can be considered. FIG. 13 shows a schematic diagram. In the method shown in FIG. 13, the characteristic of each POI is assigned to the nearest road. However, this method does not properly consider the relationship between the road and the POI, and cannot accurately estimate the characteristics.

  The characteristic of POI is considered to be a combination of a part affected by the road in contact with the POI and a part unique to POI. For example, a restaurant near a busy road is thought to be a lively atmosphere, but the characteristics of what food is served are irrelevant to the road.

  Moreover, it is considered that the characteristic that the POI is influenced by the road also affects other POIs that are in contact with the same road. Without these considerations, it is impossible to estimate the characteristics of the road itself. It is also necessary to perform estimation in consideration of the road structural relationship. This is because there are roads for which data on characteristics cannot be obtained as shown in FIG.

  In addition, as a method for understanding the characteristics of a road by using characteristic understanding in units of areas, a technique for imparting the characteristics to a road in the area can be considered. FIG. 14 shows a schematic diagram. However, with this method, the characteristics of the roads in the area are uniform, and there is a problem that, for example, in route search, it is not known which of the roads in the area meets the user's needs.

  The present invention has been made in view of the above circumstances, and takes into account the relationship between the spot and the road and the relationship between the road, a characteristic understanding device that can accurately estimate the characteristic of the spot and the characteristic of the road, It is an object to provide a method and a program.

  In order to achieve the above object, the characteristic understanding device according to the first aspect of the present invention expresses a road network in the form of a graph with each road as an edge and each intersection as a node, and information on the road is given to the graph. Assigning the spot to each of the roads based on the road network and the location information of the spot, and generating a relationship between the roads for each of the roads taking into account the structural relationship of the road network A relationship structuring unit, a data set including word information associated with each spot, and the relationship between the spot and the road assigned to each of the roads obtained by the relationship structuring unit To optimize a predetermined objective function including the characteristics of each word included in the data set, the characteristics of each spot, and the characteristics of each road. Wherein the characteristics of each word included in the data set, is configured to include the characteristics of each spot, and the estimation unit for estimating the characteristics of each road, a.

  A characteristic understanding method according to a second invention is a characteristic understanding method in a characteristic understanding device including a relation structuring unit and a characteristic estimation unit, wherein the relation structuring unit uses each road as an edge and each intersection as a node. The spot is assigned to each of the roads based on the road network in which the network is expressed in the form of a graph and the road information is given to the graph and the position information of the spots. For each, generating a relationship between the roads considering the structural relationship of the road network, the characteristic estimation unit includes a data set including word information associated with each spot, and the relationship structuring unit Each unit included in the word information of the data set based on the spot assigned to each of the roads and the relationship between the roads obtained by Characteristics of each word included in the data set, characteristics of each spot, and characteristics of each road so as to optimize a predetermined objective function including characteristics of each of the spots, characteristics of each spot, and characteristics of each road Estimating the characteristics of

  According to the first and second inventions, the relation structuring unit expresses the road network in the form of a graph with each road as an edge and each intersection as a node, and road information is given to the graph. And a spot is assigned to each road based on the position information of the spot, and a relation between roads is generated for each road in consideration of the structural relation of the road network.

  And based on the data set including the word information associated with each spot and the relationship between the spot and the road allocated to each of the roads obtained by the relationship structuring unit, the characteristic estimation unit , Characteristics of each word included in the data set so as to optimize a predetermined objective function including the characteristics of each word included in the word information of the data set, the characteristics of each spot, and the characteristics of each road And the characteristics of each spot and the characteristics of each road.

  In this way, a spot is assigned to each road based on the road network and the position information of the spot, and for each road, a relation between the roads considering the structural relationship of the road network is generated. Based on the data set including the word information linked to and the relationship between the spot and the road assigned to each road, the characteristics of each word included in the data set, the characteristics of each spot, By estimating the characteristics of each word, the characteristics of each spot, and the characteristics of each road so as to optimize a predetermined objective function including the characteristics of each road, the relationship between spots and roads and Considering the relationship between roads, it is possible to accurately estimate spot characteristics and road characteristics.

  The objective function of the present invention includes the number of times each word represented by the word information of the data set associated with each spot appears, the characteristics of each word included in the word information of the data set, and the characteristics of each spot And the likelihood of the data set for each spot calculated using the generation probability that each word is generated in the spot assigned to the road, represented using the characteristics of each road, and the road And a constraint on the characteristics of each road expressed by using the similarity of the characteristics of the road between the connected roads obtained from the relationship between the roads.

  In addition, the present invention provides a word characteristic storage unit that stores the characteristics of each word included in the word information of the data set estimated by the characteristic estimation unit, and the characteristics of each spot estimated by the characteristic estimation unit. In the position characteristic storage unit, the word preprocessing unit that converts the input word into the word characteristic based on the characteristics of each word stored in the word characteristic storage unit, and the position characteristic storage unit A fitness calculation for calculating a fitness between the input word and each of the spots based on the stored characteristics of each spot and the characteristics of the input word converted by the word preprocessing unit. And a portion.

  In addition, the present invention provides a word characteristic storage unit that stores characteristics of each word included in the word information of the data set estimated by the characteristic estimation unit, and a characteristic of each road estimated by the characteristic estimation unit. In the road characteristic storage unit, the word preprocessing unit that converts the input word into the word characteristic based on the characteristic of each word stored in the word characteristic storage unit, and the road characteristic storage unit A fitness calculation for calculating a fitness between the input word and each of the roads based on the stored characteristics of each road and the characteristics of the input word converted by the word preprocessing unit. And a portion.

  Moreover, the program of this invention is a program for functioning a computer as each part which comprises the said characteristic understanding apparatus.

  As described above, according to the characteristic understanding device, method, and program of the present invention, a spot is assigned to each road based on the road network and the position information of the spot, Based on the data set including word information associated with each spot and the relationship between the spot and the road assigned to each of the roads, generating the relationship between the roads considering the structural relationship of the network, Each word characteristic, each spot characteristic, and each so as to optimize a predetermined objective function including characteristics of each word included in the data set, characteristics of each spot, and characteristics of each road By estimating the road characteristics, it is possible to accurately estimate the spot characteristics and the road characteristics in consideration of the relationship between the spot and the road and the relationship between the roads. It is obtained.

It is a figure for demonstrating the outline | summary of this Embodiment. It is a block diagram which shows the functional structure of the characteristic understanding apparatus which concerns on this Embodiment. It is a figure which shows an example of POI data. It is a figure which shows an example which converted the format of POI data. It is a figure which shows an example of the latitude longitude information of POI. It is a figure which shows an example of a road network. It is a figure which shows an example of a road network. It is a figure which shows an example of the characteristic of a word, the characteristic of POI, and the characteristic of a road. It is a figure which shows an example of a process of the word pre-processing part. It is a figure which shows an example of the process of calculation of a fitness. It is a flowchart showing the process of the characteristic estimation process routine in the characteristic understanding apparatus which concerns on this embodiment. It is a flowchart showing the process of the compatibility calculation process routine in the characteristic understanding apparatus which concerns on this embodiment. It is a figure which shows an example of a prior art. It is a figure which shows an example of a prior art.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

<Outline of Embodiment of the Present Invention>

  First, an outline of an embodiment of the present invention will be described. Embodiments of the present invention relate to a technique for understanding the real world situation based on information on the Web such as social media. In the present embodiment, a point of interest (POI), which is an example of a spot, is used as a data set unit. POI is a spot that people visit with interest, such as stores and sightseeing spots.

  In the embodiment of the present invention, the characteristics of the road are estimated in consideration of the relationship between the POI and the road and the structural relationship between the roads, and the characteristic of the POI is also estimated at the same time. In addition, in order for a person to find a POI or road having a desired characteristic, a characteristic of a word that can be collated with the characteristic of the POI or the characteristic of the road is required, so the characteristic of the word is also estimated at the same time.

  FIG. 1 is a diagram for explaining the present embodiment.

  First, how to consider the relationship between POI and road will be described. As shown in FIG. 1, the POI characteristic is considered to be a combination of a part affected by the road that the POI touches and a part unique to the POI, and in this embodiment, they are estimated simultaneously. Think of a mechanism that can do this.

  Next, the structural relationship between roads will be described. As shown in FIG. 1 above, road characteristics may have characteristics such as connected roads having similar characteristics, and roads that are connected but of different scales have reduced characteristics. It is done.

  For example, when two roads are similar in scale and connected straight, it is considered that the atmosphere is close and the characteristics are similar. On the other hand, if the connection is not straight but 90 degrees, the similarity of characteristics is considered to decrease.

  Considering the above two points, the common part is extracted as a road characteristic for roads with many POIs, and the characteristics of connected roads are used for roads with insufficient POI data. This makes it possible to estimate the road characteristics with high accuracy.

  Further, by using a combination of the characteristics of the word, POI, and road obtained as a result of the above estimation, a numerical value indicating how much each POI and each road is applied to the keyword for a given keyword. It is possible to calculate the degree of fitness that is. By using this, it is possible to perform a route search by specifying various keywords, a route search that passes in front of a POI suitable for the keyword, and the like.

<Configuration of Character Understanding Device According to Embodiment of the Present Invention>
Next, the configuration of the characteristic understanding device according to the embodiment of the present invention will be described. As shown in FIG. 2, the characteristic understanding device 100 according to the embodiment of the present invention stores a CPU, a RAM, a program for executing a characteristic estimation processing routine and a fitness calculation processing routine described later, and various data. And a computer including a ROM. Functionally, the characteristic understanding device 100 includes an input unit 10, a calculation unit 20, and an output unit 50 as shown in FIG. The characteristic understanding device according to the present embodiment uses the POI data and the road network to estimate the characteristics of each word, the characteristics of each POI, and the characteristics of each road. In addition, the characteristic understanding device according to the present embodiment uses the output results regarding the characteristics of each word, the characteristics of each POI, and the characteristics of each road, and the matching degree between the word given from the outside and each road, The degree of matching between the word and each POI is output.

  The input unit 10 receives a word as an input from the outside.

  The calculation unit 20 includes a POI data storage unit 22, a POI data preprocessing unit 24, a POI latitude / longitude storage unit 26, a road network storage unit 28, a road POI relationship structuring unit 30, a characteristic estimation unit 32, A word characteristic storage unit 40, a position characteristic storage unit 42, a road characteristic storage unit 44, a word preprocessing unit 46, and a fitness calculation unit 48 are configured.

  The POI data storage unit 22 stores text data for each POI. Specifically, the POI data storage unit 22 stores data related to each POI (hereinafter referred to as POI data) such as data collected from information on the Web such as social media. An example of POI data is text data posted for each POI, for example. FIG. 3 shows an example of POI data.

  The POI data preprocessing unit 24 processes the data for each of the POI data stored in the POI data storage unit 22 into a format suitable for the processing of the characteristic estimation unit 32 described later, and includes a data set including the POI and word information. Outputs spot information linked with. For example, the POI data preprocessing unit 24 is a data format in which the count of how many times each word appears in the text from the text of the POI data, and is an example of a data set including word information, Bag of Words. Is generated. FIG. 4 shows an example in which the format of the POI data is changed by the POI data preprocessing unit 24. In the example of FIG. 4, it is assumed that the POI data to be input is the POI data of FIG. Further, information in which the POI data pre-processing unit 24 has changed the format of the POI data is hereinafter referred to as spot information. As shown in the example of FIG. 4, the spot information is obtained by associating POI and a data set including word information.

  The POI latitude / longitude storage unit 26 stores latitude / longitude information (hereinafter, POI position information) for each POI. FIG. 5 shows an example of POI position information.

  The road network storage unit 28 stores a road network. A road network is an expression of data in a format in which road information is expressed in a graph format with each road as an edge and each intersection as a node, and road information representing metadata such as the road scale is given to the graph. That is.

  As an example of a specific data format, one in which each road is expressed as a list of coordinate point data having a plurality of latitudes and longitudes can be considered. In this case, the coordinate points of the start point and end point of each road become nodes in the road network, and roads having a common node as the start point or end point are connected. In addition, metadata such as road scale is stored for each road. An example of a road network is shown in FIG.

  Based on the road network stored in the road network storage unit 28 and each POI position information stored in the POI latitude / longitude storage unit 26, the relationship structuring unit 30 calculates the POI for each road. assign. Further, the relationship structuring unit 30 generates a graph as a relationship between roads in consideration of the structural relationship of the road network for each road based on the road network stored in the road network storage unit 28.

  Specifically, the relationship structuring unit 30 reads data from the POI latitude / longitude storage unit 26 and the road network storage unit 28 and outputs two tables. The first is a table in which each POI is assigned to the nearest road. The second is a table representing a graph created from the road network. In this graph, roads are nodes, edges are connected between connected roads, and weights are set according to the relationship between the roads. This weight represents how similar the characteristics are considered between roads. For example, the following realization methods (1) to (4) are conceivable.

(1) When roads are connected, the initial value of the weight between roads is set to 1.
(2) If the scale is different, multiply by 0.3.
(3) If there is a large road at the same intersection, multiply by 0.3.
(4) Assuming that the angle between the roads is 180 degrees and 1 and 90 degrees is 0.3, a value continuously changed according to the angle between the roads is applied.

  An input / output example in the relation structuring unit 30 is shown in FIG. The data actually input to the relationship structuring unit 30 is the data shown in FIG. 5 and FIG. 6 above, but the roads and POIs visualized on the latitude and longitude coordinates are input in FIG. An example is shown.

  In the input example shown in FIG. 7, POI1 and POI2 have road 1 as the nearest neighbor. Therefore, the road 1 is linked to POI1 and POI2 in the relationship table between the road and POI in the output. Since road 1 is connected to road 2 and road 3, edges are set for road 1-road 2 and road 1-road 3, respectively, and output as a table representing a graph reflecting the road network. The

  Moreover, since the road 1 and the road 2 are connected straight, it is considered that the similarity of characteristics is high, and thus the weight is large. On the other hand, since the road 1 and the road 3 are not straight and the characteristic similarity is considered to be low, the weight is small.

  The characteristic estimation unit 32 includes a data set including word information for each POI obtained by the POI data preprocessing unit 24, and the POI and graph assigned to each road obtained by the relationship structuring unit 30. The word information of the data set so as to optimize a predetermined objective function including the characteristics of each word included in the word information of the data set, the characteristics of each POI, and the characteristics of each road. The characteristics of each word, the characteristics of each POI, and the characteristics of each road are estimated. The characteristic estimation unit 32 includes a word characteristic estimation unit 34, a POI characteristic estimation unit 36, and a road characteristic estimation unit 38.

  The characteristic estimation unit 32 receives the outputs of the POI data preprocessing unit 24 and the road POI relationship structuring unit 30, and simultaneously estimates the characteristics of words, POIs, and roads. At that time, the relationship between the POI and the road and the structure of the road are considered.

  As a method of realizing this function, see pLSA (Probabilistic Latent Semantic Analysis) (for example, T. Hofmann. “Probabilistic latent semantic indexing.” In Proc. Of SIGIR, 1999) considering graph constraints and hierarchical structure. ) May be used. pLSA is a technology that receives Bag of Words of each document from text data and outputs a vector that expresses the meaning of each word and each document. When pLSA is used as a processing method, it can be realized by considering the text of each POI as a document and extending it so as to consider the hierarchical structure of the road and its neighboring POIs and the structure of the road network.

  When considering the hierarchical structure, for example, use Sparse Additive Generative Model (for example, refer to J. Eisenstein et al, “Sparse additive generative models of text” in Proc. Of ICML, 2011). Can be realized.

  Next, specific formulas for the case where text is used as information will be shown. A set of all text data is defined as C, a set of roads r is defined as a node set R, an edge is defined between connected roads as E, and G is represented as G = (R, E). When a graph is used, the overall objective function is defined as in the following equation (1), and the characteristics of the word, POI, and road are estimated so as to minimize the objective function.

(1)

  Here, L (C, G) represents the likelihood of each POI data set calculated from the generation probability of each word. R (G) represents a restriction on the characteristics of each road based on the graph structure. λ is a weight term related to these two terms.

  The set of POIs near the road r is Pr, the number of occurrences of the word w represented by the word information of the data set linked to each POIp is c (w, p), and the word w is the POIp near the road r. When the generation probability to be generated is p (w | r, p), L (C, G) is defined as follows.

(2)

  Further, p (w | r, p) is defined as follows.

(3)

  As shown in the above formula (3), the generation probability p (w | r, p) at which each word w is generated in the POIp assigned to the road r is the characteristic of each word included in the text data, the POI of each POI It is expressed using the characteristics and the characteristics of each road.

Here, ψ is a vector representing the characteristics of each road, θ is a vector representing the characteristics of each POI, and φ represents the word distribution of each topic. ψ ^r _z represents the size of the z-th dimension of the vector representing the characteristics of the road r, and θ ^p _z represents the size of the z-th dimension of the vector representing the characteristics of POIp. φ ^z _w represents the generation probability of word w in topic z. K represents the number of topics, and V represents a set of vocabularies that appear in the entire text.

The above formula will be described in detail. The generation probability of each word is calculated by multiplying the generation probability of a certain topic z in the POI by the generation probability of the word in the topic z for all topics and taking the sum. The topic generation probability in a certain POIp is calculated from a value obtained by adding the difference between POIp and the road to ψ ^r _z which is a parameter indicating the ease of generating the topic z in the road r adjacent to the POI. . By calculating p (w | r, p) using this equation, there are words that are likely to be generated commonly in the vicinity POI of a road r, and there are also words that are likely to be generated individually at each POI. Can be expressed.

  Further, R (G) in the above formula (1) is defined as follows.

(4)

Here, w (r, r ′) represents the strength of the similarity between the characteristics of the roads r and r ′. Details will be described later. diff (ψ ^r , ψ ^{r ′} ) is a function for calculating the degree of coincidence of the characteristic ψ, and can be realized by an inner product or the like. By maximizing this formula, the estimated value of each characteristic can be made close to each other on a road considered to have a high degree of characteristic similarity.

w (r, r ′) will be described. w (r, r ′) represents the similarity of characteristics between roads, and is a weight output from the road POI relationship structuring unit 30. Using this, it is possible to obtain ψ while taking into account the similarity of characteristics coming from the structural relationship of the road.
Therefore, R (G) in the above equation (1) represents the similarity diff (ψ ^r , ψ ^{r ′} ) of the road characteristics between the connected roads obtained from the graph output from the relation structuring unit 30. It is expressed using.

  Considering the above, the characteristics of the word, POI, and road are estimated by optimizing φ, ψ, and θ. Gibbs Sampling can be used as an optimization method. In this embodiment, a case where φ, ψ, and θ are estimated using Gibbs sampling will be described as an example.

  Specifically, first, φ, ψ, and θ are initialized appropriately.

  Next, the word characteristic estimation unit 34 fixes ψ and θ that are initialized or estimated last time, and optimizes φ with respect to the above equation (1).

  The POI characteristic estimator 36 optimizes ψ with respect to the above equation (1) by fixing θ initialized or previously estimated and φ estimated by the word characteristic estimator 34.

  The road characteristic estimator 38 fixes φ estimated by the word characteristic estimator 34 and ψ estimated by the POI characteristic estimator 36, and optimizes θ for the above equation (1).

  Then, by repeating the estimation processing of the word characteristic estimation unit 34, the POI characteristic estimation unit 36, and the road characteristic estimation unit 38 until a predetermined condition is satisfied, the objective function of the above equation (1) is optimized. it can.

  The optimization in the word characteristic estimation unit 34, the POI characteristic estimation unit 36, and the road characteristic estimation unit 38 is performed using Projected Scaled Sub-Gradient (for example, references (M. Schmidt, A. Niculescu-Mizil, and K. Murphy, “ Learning graphical model structure using l1-regularization paths ”, in Proc. Of AAAI, 2007.)). FIG. 8 shows an example of output.

  The word characteristic storage unit 40 stores the characteristics of each word estimated by the characteristic estimation unit 32.

  The position characteristic storage unit 42 stores the characteristics of each POI estimated by the characteristic estimation unit 32.

  The road characteristic storage unit 44 stores the characteristics of each road estimated by the characteristic estimation unit 32.

  FIG. 8 shows an example of word characteristics stored in the word characteristic storage unit 40, an example of POI characteristics stored in the position characteristic storage unit 42, and an example of road characteristics stored in the road characteristic storage unit 44. .

  The word preprocessing unit 46 converts the word input from the input unit 10 into a word characteristic based on the characteristic of each word stored in the word characteristic storage unit 40. Specifically, the word preprocessing unit 46 has received the characteristic of the word that matches the word among the characteristics of each word stored in the word characteristic storage unit 40 for the word received in the input unit 10. Get as a characteristic of the word. FIG. 9 shows an example of the acquired word characteristics when the accepted word is “coffee shop”.

  The matching degree calculation unit 48 matches the inputted word and each POI based on the characteristics of each POI stored in the position characteristic storage unit 42 and the characteristics of the word converted by the word preprocessing unit 46. The degree is calculated, and the result is output from the output unit 50.

  In addition, the fitness level calculation unit 48 determines the input word and each road based on the characteristics of each road stored in the road characteristic storage unit 44 and the characteristics of the words converted by the word preprocessing unit 46. And the result is output from the output unit 50.

  In the present embodiment, since the characteristics of each word, the characteristics of each POI, and the characteristics of each road are expressed as vectors, the degree of matching between words and POI and the degree of matching between words and roads are Calculate with the dot product of the vector of properties. FIG. 10 shows an example of input and output data in the fitness level calculation unit 48.

  The output unit 50 receives the output of the fitness calculation unit 48 and outputs it to the outside.

<Operation of the characteristic understanding device according to the embodiment of the present invention>
Next, the operation of the characteristic understanding device 100 according to the embodiment of the present invention will be described. In the embodiment of the present invention, as a processing process, a processing flow for estimating and accumulating each characteristic as shown in FIG. 11, and a process for calculating the fitness between the input word and each characteristic as shown in FIG. There is a flow.

  First, when the characteristic understanding device 100 reads each POI data from the POI data storage unit 22, the characteristic understanding apparatus 100 executes a characteristic estimation processing routine shown in FIG.

  First, in step S100, the POI data preprocessing unit 24 processes the data for each of the read POI data into a format suitable for the processing in step S102, which will be described later, and links the POI and a data set including word information to each other. Get information.

  Next, in step S102, the relationship structuring unit 30 is based on the road network stored in the road network storage unit 28 and each POI position information stored in the POI latitude / longitude storage unit 26. Assign a POI to each of the roads.

  In step S104, the characteristic estimation unit 32 includes the data set including the word information for each POI obtained in step S100, the POI and graph assigned to each of the roads obtained in step S102. Based on the above, the characteristics of each word included in the data set, the characteristics of each POI, and the characteristics of each road are estimated so as to optimize the objective function shown in the above equation (1).

  In step S106, the characteristic estimation unit 32 stores the characteristics of each word estimated in step S104 in the word characteristic storage unit 40, stores the characteristics of each POI in the position characteristic storage unit 42, and stores the characteristics of each road. The characteristic is stored in the road characteristic storage unit 44, and the characteristic estimation processing routine is terminated.

  Next, when receiving the word at the input unit 10, the characteristic understanding device 100 executes a fitness level calculation processing routine shown in FIG.

  First, in step S200, the characteristics of each word stored in the word characteristic storage unit 40, the characteristics of each POI stored in the position characteristic storage unit 42, and the roads stored in the road characteristic storage unit 44 And read the characteristics.

  Next, in step S202, the word preprocessing unit 46 acquires the characteristics of the word received by the input unit 10 based on the characteristics of each word acquired in step S200.

  Next, in step S204, the fitness calculation unit 48 determines the word and each POI based on the characteristics of each POI acquired in step S200 and the characteristics of the word received in the input unit 10 acquired in step S202. Each of the goodness of fits is calculated. Further, the goodness-of-fit calculation unit 48 determines each of the goodness of fit between the word and each road based on the characteristics of each road acquired in step S200 and the characteristics of the word received in the input unit 10 acquired in step S202. Is calculated.

  Next, in step S206, each of the fitness of the word received in the input unit 10 acquired in step S204 and each POI, and each of the fitness of the word received in the input unit 10 and each road are output to the output unit. 50, and the fitness calculation processing routine is terminated.

  As described above, according to the characteristic understanding device according to the embodiment of the present invention, a POI is assigned to each road based on the road network and POI position information, and the road network is assigned to each road. Data is generated based on a data set including word information associated with each POI and a POI assigned to each road and a relation between the roads. The characteristics of each word, the characteristics of each POI, and each road so as to optimize a predetermined objective function including the characteristics of each word included in the set, the characteristics of each POI, and the characteristics of each road Thus, it is possible to accurately estimate the POI characteristic and the road characteristic in consideration of the relationship between the POI and the road and the relationship between the roads.

  Also, based on the characteristics of each stored word, the input word is converted into characteristics, and on the basis of the stored characteristics of each POI and the characteristics of each road, and the characteristics of the converted input word By calculating the degree of matching between the inputted word and each of the POIs and the degree of fitting between the inputted words and each of the roads, the relation between the POI and the roads and the relation between the roads is considered. The degree of matching with each POI and the degree of matching between each word and each road can be calculated.

  Also, using the information that exists on the Web, the degree of compatibility of various POIs and roads with various keywords is calculated with high accuracy while taking into account the relation between roads and POIs and the relations between roads. Can be built. As a result, for various keywords such as those related to human subjectivity, a service such as a route search focusing on a POI or road that matches the keyword can be realized.

  In addition, it is possible to understand real-world information based on information on the Web such as social media.

  Note that the present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the gist of the present invention.

  For example, in the present embodiment, the case of calculating the degree of matching between the input word and each POI and the degree of matching between the input word and each road has been described as an example, but the present invention is not limited to this. Instead, only one of the degree of matching between the word and each POI and the degree of matching between the word and each road may be calculated.

  In the present embodiment, the case where the POI-related data is text data has been described. However, the present invention is not limited to this, and the POI-related data includes a word set added as metadata to image data. It may be used. In this case, each process of the characteristic understanding device according to the present embodiment may be performed by regarding one image data for each POI as one text and regarding a set of words included in the image data as the above-described one text.

  In the present embodiment, the case where pLSI is used to estimate the characteristics of words, POIs, and roads has been described. However, the present invention is not limited to this, and other methods are used. Also good.

  In this embodiment, the case where POI data and POI position information are stored separately has been described, but the present invention is not limited to this. For example, POI data and POI position information may be combined and stored as POI data. In this case, in this embodiment, the POI position information is acquired from the POI data storage unit 22.

  Moreover, although the case where the unit of the data set is POI has been described as an example, the present invention is not limited to this. For example, a point or region other than the POI may be used as a unit of the data set.

  In addition, the above-described characteristic understanding device 100 has a computer system therein, but the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used. .

  Further, in the present specification, the embodiment has been described in which the program is installed in advance. However, the program can be provided by being stored in a computer-readable recording medium such as a CD-ROM or a memory card. It is.

DESCRIPTION OF SYMBOLS 10 Input part 20 Calculation part 22 Data storage part 24 POI data pre-processing part 26 Latitude / longitude storage part 28 Road network storage part 30 Relation structuring part 32 Characteristic estimation part 34 Word characteristic estimation part 36 Characteristic estimation part 38 Road characteristic estimation part 40 Word characteristic storage unit 42 Position characteristic storage unit 44 Road characteristic storage unit 46 Word preprocessing unit 48 Fitness calculation unit 50 Output unit 100 Character understanding device

Claims (8)

Each road is represented based on a road network in which each road is an edge and each intersection is a node in the form of a graph, and the road information is attached to the graph and the spot position information. And assigning the spots to each of the roads, and for each of the roads, a relation structuring unit that generates a relation between the roads in consideration of a structural relation of the road network
Based on a data set including word information associated with each spot, and the relationship between the spot and the road assigned to each of the roads obtained by the relation structuring unit, Characteristics of each word included in the data set, so as to optimize a predetermined objective function including characteristics of each word included in the data set, characteristics of each spot, and characteristics of each road, A characteristic estimation unit for estimating the characteristics of the spot and the characteristics of each road;
Including device for understanding characteristics. The objective function includes the number of times each word represented by the word information of the data set associated with each spot appears, the characteristics of each word included in the word information of the data set, the characteristics of each spot, and each The likelihood of the data set of each spot calculated using the generation probability that each word is generated in the spot assigned to the road, expressed using the characteristics of the road, and the relationship between the roads The characteristic understanding device according to claim 1, further comprising: a restriction on a characteristic of each road represented by using a similarity of the characteristic of the road between the connected roads obtained from A word characteristic storage unit that stores characteristics of each word included in the word information of the data set estimated by the characteristic estimation unit;
A position characteristic storage unit that stores the characteristics of each spot estimated by the characteristic estimation unit;
A word pre-processing unit that converts an input word into a characteristic of the word based on characteristics of each word stored in the word characteristic storage unit;
Based on the characteristics of each spot stored in the position characteristic storage unit and the characteristics of the input word converted by the word pre-processing unit, the degree of matching between the input word and each of the spots A fitness calculation unit for calculating
The characteristic understanding device according to claim 1 or 2, further comprising: A word characteristic storage unit that stores characteristics of each word included in the word information of the data set estimated by the characteristic estimation unit;
A road characteristic storage unit that stores the characteristics of each road estimated by the characteristic estimation unit;
A word pre-processing unit that converts an input word into a characteristic of the word based on characteristics of each word stored in the word characteristic storage unit;
Based on the characteristics of each road stored in the road characteristic storage unit and the characteristics of the input word converted by the word pre-processing unit, the degree of fitness between the input word and each of the roads A fitness calculation unit for calculating
The characteristic understanding device according to claim 1 or 2, further comprising: A characteristic understanding method in a characteristic understanding device including a relation structuring unit and a characteristic estimation unit,
The relation structuring unit expresses the road network in the form of a graph with each road as an edge and each intersection as a node, and road information in which the road information is attached to the graph, and spot position information. And assigning the spots to each of the roads, and generating, for each of the roads, a relationship between the roads that takes into account the structural relationship of the road network;
The property estimation unit is a data set including word information associated with each spot, and the relationship between the spot and the road assigned to each of the roads obtained by the relationship structuring unit. To optimize the predetermined objective function including the characteristics of each word included in the word information of the data set, the characteristics of each spot, and the characteristics of each road. Estimating the characteristics of each included word, the characteristics of each spot, and the characteristics of each road;
Characteristic understanding method including. A word characteristic storage unit storing the characteristics of each word included in the word information of the data set estimated by the characteristic estimation unit, and a position characteristic storage storing the characteristics of each spot estimated by the characteristic estimation unit A word pre-processing unit, and a fitness calculation unit,
The word preprocessing unit, based on the characteristics of each word stored in the word characteristic storage unit, to convert the input word into the characteristics of the word;
Based on the characteristics of each spot stored in the position characteristic storage unit and the characteristics of the input word converted by the word preprocessing unit, the fitness level calculation unit calculates the input word and the Calculating the fitness with each of the spots;
The method for understanding characteristics according to claim 5, comprising: A word characteristic storage unit storing characteristics of each word included in the word information of the data set estimated by the characteristic estimation unit; and a road characteristic storage storing characteristics of each road estimated by the characteristic estimation unit. A word pre-processing unit, and a fitness calculation unit,
The word preprocessing unit, based on the characteristics of each word stored in the word characteristic storage unit, to convert the input word into the characteristics of the word;
Based on the characteristics of each road stored in the road characteristic storage unit and the characteristics of the input word converted by the word preprocessing unit, the fitness calculation unit is configured to input the input word and the Calculating the fitness with each of the roads;
The method for understanding characteristics according to claim 5, comprising:   The program for functioning a computer as each part which comprises the characteristic understanding apparatus of any one of Claims 1-4.